Medical question answering benchmark derived from the US Medical Licensing Examination (USMLE). 1,273 multiple-choice questions covering clinical knowledge across 21 medical specialties. Scraped from Vals.ai.
Professional-level medical knowledge; ground-truth answers from real licensing exams; widely cited in medical AI literature.
Multiple-choice format limits measurement of reasoning; approaching saturation on frontier models (~95%+); US-centric curriculum; public dataset with contamination risk.
Higher is better. Percentage of USMLE-style questions answered correctly.
Multiple-choice format limits measurement of reasoning.
| # | Model | Score |
|---|---|---|
| 1 | 97.5 | |
| 2 | 97.4 | |
| 3 | 97.3 | |
| 4 | 96.8 | |
| 5 | 96.8 | |
| 6 | 96.8 | |
| 7 | 96.8 | |
| 8 | 96.7 | |
| 9 | 96.7 | |
| 10 | 96.6 | |
| 11 | 96.5 | |
| 12 | 96.5 | |
| 13 | 96.4 | |
| 14 | 96.4 | |
| 15 | 96.4 | |
| 16 | 96.3 | |
| 17 | 96.3 | |
| 18 | 96.3 | |
| 19 | 96.2 | |
| 20 | 96.1 | |
| 21 | 96.1 | |
| 22 | 96.1 | |
| 23 | 96.1 | |
| 24 | 96.1 | |
| 25 | 96.0 | |
| 26 | 96.0 | |
| 27 | 96.0 | |
| 28 | 95.9 | |
| 29 | 95.9 | |
| 30 | 95.9 | |
| 31 | 95.9 | |
| 32 | 95.8 | |
| 33 | 95.8 | |
| 34 | 95.8 | |
| 35 | 95.7 | |
| 36 | 95.7 | |
| 37 | 95.7 | |
| 38 | 95.6 | |
| 39 | 95.6 | |
| 40 | 95.6 | |
| 41 | 95.6 | |
| 42 | 95.5 | |
| 43 | 95.4 | |
| 44 | 95.4 | |
| 45 | 95.4 | |
| 46 | 95.4 | |
| 47 | 95.3 | |
| 48 | 95.2 | |
| 49 | 95.1 | |
| 50 | 95.0 |